Content
75%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A strong, lean instruction-only skill: executable gh commands, exact output schema, tiered scoring criteria, and a well-sequenced analysis process with override and calibration checkpoints. The main weaknesses are the unspecified composite-score weights (hurting reproducibility of the headline number), a redundant Role & Purpose section, and a 185-line body whose detailed scoring tiers could be split into a reference file.
Suggestions
Specify the explicit dimension weights for the composite risk_score (e.g., data_integrity 0.30, security_surface 0.25, others summing the rest) instead of only stating 'Data integrity and security carry more weight', so the headline numeric output is reproducible across runs.
Trim the 'Role & Purpose' section, which nearly restates the frontmatter description, keeping at most the one-line calibration framing.
Either move the Data Integrity scoring tiers (and optionally the per-area path classification) into a references/ file linked from the body, or add a final validation step before returning (all six dimensions present, overrides and calibration safeguard applied, output is valid JSON).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence — no explanation of what a PR or `gh` is, no library comparisons — with every scoring rule earning its place. It falls just short of anchor 5 because of minor trimmable material: the 'Role & Purpose' paragraph nearly verbatim restates the frontmatter description, and 'Think like a senior QA engineer: what would keep you up at night?' is flavor. Not anchor 3, since there is no genuinely unnecessary explanation, only these minor instances. | 4 / 5 |
Actionability | Guidance is mostly executable: copy-paste-ready `gh pr view --json …` and `gh pr diff` commands, exact categorical thresholds, tiered data-integrity criteria, and a complete output JSON schema. It misses anchor 5 on one real gap: the core numeric output is 'a weighted overall score' where 'Data integrity and security carry more weight' — the actual dimension weights are never given, so the executor must invent them and results are not reproducible. Not anchor 3, since nothing is pseudocode or high-level hand-waving; the gap is specific and minor relative to the whole. | 4 / 5 |
Workflow Clarity | A clear, correctly ordered 7-step process with an explicit pre-compute stage ('Blast Radius (pre-compute before scoring)') and substantive checkpoints: hard HIGH overrides, a calibration safeguard ('regression_surface ≥ 5 AND complexity ≥ 4 AND data_integrity ≤ 2 → MEDIUM'), and a truncated-diff fallback rule. It misses anchor 5 because there is no final validation of the emitted result (e.g., verifying all six dimensions are present, overrides were applied, and the JSON is well-formed before returning). Not anchor 3, since checkpoints are explicit, not merely implied. | 4 / 5 |
Progressive Disclosure | Scored against the actual bundle (SKILL.md only, no references/scripts/assets): sections are well-organized, self-contained, and nothing is buried or nested. It does not reach anchor 5 because that anchor (and the under-50-line simple-skill exception) doesn't apply — at ~185 lines, the ~30-line Data Integrity scoring tiers and the detailed analysis heuristics are candidates for a one-level-deep reference file. Not anchor 3, since the structure is good and all present content is appropriately placed; only a minor split opportunity exists. | 4 / 5 |
Total | 16 / 20 Passed |